Toward Addiction Prediction: An Overview of Cross-Validated Predictive Modeling Findings and Considerations for Future Neuroimaging Research

Prediction of craving across studies: A commentary on conceptual and methodological considerations when using data-driven methods

Craving is a central feature of substance use disorders and disorders due to addictive behaviors. Considerable research has investigated neural mechanisms involved in the development and processing of craving. Recently, connectome-based predictive modeling, a data-driven method, has been used in four studies aiming to predict craving related to substance use, addictive behaviors, and food. Studies differed in methods, samples, and conceptualizations of craving. Within the commentary we aim to compare, contrast and consolidate findings across studies by considering conceptual and methodological features of the studies. We derive a theoretical model on the functional connectivity-craving relationships across studies.

Sex-specific differences in brain activity dynamics of youth with a family history of substance use disorder

An individual's risk of substance use disorder (SUD) is shaped by a complex interplay of potent biosocial factors. Current neurodevelopmental models posit vulnerability to SUD in youth is due to an overreactive reward system and reduced inhibitory control. Having a family history of SUD is a particularly strong risk factor, yet few studies have explored its impact on brain function and structure prior to substance exposure. Herein, we utilized a network control theory approach to quantify sex-specific differences in brain activity dynamics in youth with and without a family history of SUD, drawn from a large cohort of substance-naïve youth from the Adolescent Brain Cognitive Development Study. We summarize brain dynamics by calculating transition energy, which probes the ease with which a whole brain, region or network drives the brain towards a specific spatial pattern of activation (i.e., brain state). Our findings reveal that a family history of SUD is associated with alterations in the brain's dynamics wherein: i) independent of sex, certain regions' transition energies are higher in those with a family history of SUD and ii) there exist sex-specific differences in SUD family history groups at multiple levels of transition energy (global, network, and regional). Family history-by-sex effects reveal that energetic demand is increased in females with a family history of SUD and decreased in males with a family history of SUD, compared to their same-sex counterparts with no SUD family history. Specifically, we localize these effects to higher energetic demands of the default mode network in females with a family history of SUD and lower energetic demands of attention networks in males with a family history of SUD. These results suggest a family history of SUD may increase reward saliency in males and decrease efficiency of top-down inhibitory control in females. This work could be used to inform personalized intervention strategies that may target differing cognitive mechanisms that predispose individuals to the development of SUD.

Risk and Promotive Factors Related to Cannabis Use Among American Indian Adolescents

Reservation-dwelling American Indian adolescents are at exceedingly high risk for cannabis use. Prevention initiatives to delay the onset and escalation of use are needed. The risk and promotive factors approach to substance use prevention is a well-established framework for identifying the timing and targets for prevention initiatives. This study aimed to develop predictive models for the usage of cannabis using 22 salient risk and promotive factors. Models were developed using data from a cross-sectional study and further validated using data from a separate longitudinal study with three measurement occasions (baseline, 6-month follow-up, 1-year follow-up). Application of the model to longitudinal data showed an acceptable performance contemporaneously but waning prospective predictive utility over time. Despite the model's high specificity, the sensitivity was low, indicating an effective prediction of non-users but poor performance in correctly identifying users, particularly at the 1-year follow-up. This divergence can have significant implications. For example, a model that misclassifies future adolescent cannabis use could fail to provide necessary intervention for those at risk, leading to negative health and social consequences. Moreover, supplementary analysis points to the importance of considering change in risk and promotive factors over time.

Brain-phenotype predictions can survive across diverse real-world data

Recent work suggests that machine learning models predicting psychiatric treatment outcomes based on clinical data may fail when applied to unharmonized samples. Neuroimaging predictive models offer the opportunity to incorporate neurobiological information, which may be more robust to dataset shifts. Yet, among the minority of neuroimaging studies that undertake any form of external validation, there is a notable lack of attention to generalization across dataset-specific idiosyncrasies. Research settings, by design, remove the between-site variations that real-world and, eventually, clinical applications demand. Here, we rigorously test the ability of a range of predictive models to generalize across three diverse, unharmonized samples: the Philadelphia Neurodevelopmental Cohort (n=1291), the Healthy Brain Network (n=1110), and the Human Connectome Project in Development (n=428). These datasets have high inter-dataset heterogeneity, encompassing substantial variations in age distribution, sex, racial and ethnic minority representation, recruitment geography, clinical symptom burdens, fMRI tasks, sequences, and behavioral measures. We demonstrate that reproducible and generalizable brain-behavior associations can be realized across diverse dataset features with sample sizes in the hundreds. Results indicate the potential of functional connectivity-based predictive models to be robust despite substantial inter-dataset variability. Notably, for the HCPD and HBN datasets, the best predictions were not from training and testing in the same dataset (i.e., cross-validation) but across datasets. This result suggests that training on diverse data may improve prediction in specific cases. Overall, this work provides a critical foundation for future work evaluating the generalizability of neuroimaging predictive models in real-world scenarios and clinical settings.

Shared and distinct structural brain networks related to childhood maltreatment and social support: connectome-based predictive modeling

Childhood maltreatment (CM) has been associated with changes in structural brain connectivity even in the absence of mental illness. Social support, an important protective factor in the presence of childhood maltreatment, has been positively linked to white matter integrity. However, the shared effects of current social support and CM and their association with structural connectivity remain to be investigated. They might shed new light on the neurobiological basis of the protective mechanism of social support. Using connectome-based predictive modeling (CPM), we analyzed structural connectomes of N = 904 healthy adults derived from diffusion-weighted imaging. CPM predicts phenotypes from structural connectivity through a cross-validation scheme. Distinct and shared networks of white matter tracts predicting childhood trauma questionnaire scores and the social support questionnaire were identified. Additional analyses were applied to assess the stability of the results. CM and social support were predicted significantly from structural connectome data (all rs ≥ 0.119, all ps ≤ 0.016). Edges predicting CM and social support were inversely correlated, i.e., positively correlated with CM and negatively with social support, and vice versa, with a focus on frontal and temporal regions including the insula and superior temporal lobe. CPM reveals the predictive value of the structural connectome for CM and current social support. Both constructs are inversely associated with connectivity strength in several brain tracts. While this underlines the interconnectedness of these experiences, it suggests social support acts as a protective factor following adverse childhood experiences, compensating for brain network alterations. Future longitudinal studies should focus on putative moderating mechanisms buffering these adverse experiences.

Brain Networks and Adolescent Alcohol Use

Importance

Alcohol misuse in adolescence is a leading cause of disability and mortality in youth and is associated with higher risk for alcohol use disorder. Brain mechanisms underlying risk of alcohol misuse may inform prevention and intervention efforts.

Objective

To identify neuromarkers of alcohol misuse using a data-driven approach, with specific consideration of neurodevelopmental sex differences.

Design, Setting, and Participants

Longitudinal multisite functional magnetic resonance imaging (fMRI) data collected at ages 14 and 19 years were used to assess whole-brain patterns of functional organization associated with current and future alcohol use risk as measured by the Alcohol Use Disorder Identification Test (AUDIT). Primary data were collected by the IMAGEN consortium, a European multisite study of adolescent neurodevelopment. Model generalizability was further tested using data acquired in a single-site study of college alcohol consumption conducted in the US. The primary sample was a developmental cohort of 1359 adolescents with neuroimaging, phenotyping, and alcohol use data. Model generalizability was further assessed in a separate cohort of 114 individuals.

Main Outcomes and Measures

Brain-behavior model accuracy, as defined by the correspondence between model-predicted and actual AUDIT scores in held-out testing data, Bonferroni corrected across the number of models run at each time point, 2-tailed α < .008, as determined via permutation testing.

Results

Among 1359 individuals in the study, the mean (SD) age was 14.42 (0.40) years, and 729 individuals (54%) were female. The data-driven, whole-brain connectivity approach identified networks associated with vulnerability for future and current AUDIT-defined alcohol use risk (primary outcome, as specified above, future: ρ, 0.22; P < .001 and present: ρ, 0.27; P < .001). Results further indicated sex divergence in the accuracies of brain-behavior models, such that female-only models consistently outperformed male-only models. Specifically, female-only models identified networks conferring vulnerability for future and current severity using data acquired during both reward and inhibitory fMRI tasks. In contrast, male-only models were successful in accurately identifying networks using data acquired during the inhibitory control-but not reward-task, indicating domain specificity of alcohol use risk networks in male adolescents only.

Conclusions and Relevance

These data suggest that interventions focusing on inhibitory control processes may be effective in combating alcohol use risk in male adolescents but that both inhibitory and reward-related processes are likely of relevance to alcohol use behaviors in female adolescents. They further identify novel networks of alcohol use risk in youth, which may be used to identify adolescents who are at risk and inform intervention efforts.

Distinct neural networks predict cocaine versus cannabis treatment outcomes

Treatment outcomes for individuals with substance use disorders (SUDs) are variable and more individualized approaches may be needed. Cross-validated, machine-learning methods are well-suited for probing neural mechanisms of treatment outcomes. Our prior work applied one such approach, connectome-based predictive modeling (CPM), to identify dissociable and substance-specific neural networks of cocaine and opioid abstinence. In Study 1, we aimed to replicate and extend prior work by testing the predictive ability of the cocaine network in an independent sample of 43 participants from a trial of cognitive-behavioral therapy for SUD, and evaluating its ability to predict cannabis abstinence. In Study 2, CPM was applied to identify an independent cannabis abstinence network. Additional participants were identified for a combined sample of 33 with cannabis-use disorder. Participants underwent fMRI scanning before and after treatment. Additional samples of 53 individuals with co-occurring cocaine and opioid-use disorders and 38 comparison subjects were used to assess substance specificity and network strength relative to participants without SUDs. Results demonstrated a second external replication of the cocaine network predicting future cocaine abstinence, however it did not generalize to cannabis abstinence. An independent CPM identified a novel cannabis abstinence network, which was (i) anatomically distinct from the cocaine network, (ii) specific for predicting cannabis abstinence, and for which (iii) network strength was significantly stronger in treatment responders relative to control particpants. Results provide further evidence for substance specificity of neural predictors of abstinence and provide insight into neural mechanisms of successful cannabis treatment, thereby identifying novel treatment targets. Clinical trials registation: "Computer-based training in cognitive-behavioral therapy web-based (Man VS Machine)", registration number: NCT01442597 . "Maximizing the Efficacy of Cognitive Behavior Therapy and Contingency Management", registration number: NCT00350649 . "Computer-Based Training in Cognitive Behavior Therapy (CBT4CBT)", registration number: NCT01406899 .

The challenges and prospects of brain-based prediction of behaviour

Relating individual brain patterns to behaviour is fundamental in system neuroscience. Recently, the predictive modelling approach has become increasingly popular, largely due to the recent availability of large open datasets and access to computational resources. This means that we can use machine learning models and interindividual differences at the brain level represented by neuroimaging features to predict interindividual differences in behavioural measures. By doing so, we could identify biomarkers and neural correlates in a data-driven fashion. Nevertheless, this budding field of neuroimaging-based predictive modelling is facing issues that may limit its potential applications. Here we review these existing challenges, as well as those that we anticipate as the field develops. We focus on the impacts of these challenges on brain-based predictions. We suggest potential solutions to address the resolvable challenges, while keeping in mind that some general and conceptual limitations may also underlie the predictive modelling approach.

Speak and you shall predict: speech at initial cocaine abstinence as a biomarker of long-term drug use behavior

Importance

Valid biomarkers that can predict longitudinal clinical outcomes at low cost are a holy grail in psychiatric research, promising to ultimately be used to optimize and tailor intervention and prevention efforts.

Objective

To determine if baseline linguistic markers in natural speech, as compared to non-speech clinical and demographic measures, can predict drug use severity measures at future sessions in initially abstinent individuals with cocaine use disorder (iCUD).

Design

A longitudinal cohort study (August 2017 - March 2020), where baseline measures were used to predict outcomes collected at three-month intervals for up to one year of follow-up.

Participants

Eighty-eight initially abstinent iCUD were studied at baseline; 57 (46 male, age 50.7+/-7.9 years) came back for at least another session.

Main Outcomes and Measures

Outcomes were self-reported symptoms of withdrawal, craving, abstinence duration and frequency of cocaine use in the past 90 days at each study session. The predictors were derived from 5-min recordings of vocal descriptions of the positive consequences of abstinence and the negative consequences of using cocaine; the baseline cocaine and other common drug use measures, demographic and neuropsychological variables were used for comparison.

Results

Models using the non-speech variables showed the best predictive performance at three(r>0.45, P<2×10-3) and six months follow-up (r>0.37, P<3×10-2). At 12 months, the natural language processing-based model showed significant correlations with withdrawal (r=0.43, P=3×10-2), craving (r=0.72, P=5×10-5), days of abstinence (r=0.76, P=1×10-5), and cocaine use in the past 90 days (r=0.61, P=2×10-3), significantly outperforming the other models for abstinence prediction.

Conclusions and Relevance

At short time intervals, maximal predictive power was obtained with models that used baseline drug use (in addition to demographic and neuropsychological) measures, potentially reflecting a slow rate of change in these measures, which could be estimated by linear functions. In contrast, short speech samples predicted longer-term changes in drug use, implying deeper penetrance by potentially capturing non-linear dynamics over longer intervals. Results suggest that, compared to the common outcome measures used in clinical trials, speech-based measures could be leveraged as better predictors of longitudinal drug use outcomes in initially abstinent iCUD, as potentially generalizable to other substance use disorders and related comorbidity.

From Computation to Clinic

Theory-driven and data-driven computational approaches to psychiatry have enormous potential for elucidating mechanism of disease and providing translational linkages between basic science findings and the clinic. These approaches have already demonstrated utility in providing clinically relevant understanding, primarily via back translation from clinic to computation, revealing how specific disorders or symptoms map onto specific computational processes. Nonetheless, forward translation, from computation to clinic, remains rare. In addition, consensus regarding specific barriers to forward translation-and on the best strategies to overcome these barriers-is limited. This perspective review brings together expert basic and computationally trained researchers and clinicians to 1) identify challenges specific to preclinical model systems and clinical translation of computational models of cognition and affect, and 2) discuss practical approaches to overcoming these challenges. In doing so, we highlight recent evidence for the ability of computational approaches to predict treatment responses in psychiatric disorders and discuss considerations for maximizing the clinical relevance of such models (e.g., via longitudinal testing) and the likelihood of stakeholder adoption (e.g., via cost-effectiveness analyses).

Candidate biomarkers in psychiatric disorders: state of the field

The field of psychiatry is hampered by a lack of robust, reliable and valid biomarkers that can aid in objectively diagnosing patients and providing individualized treatment recommendations. Here we review and critically evaluate the evidence for the most promising biomarkers in the psychiatric neuroscience literature for autism spectrum disorder, schizophrenia, anxiety disorders and post-traumatic stress disorder, major depression and bipolar disorder, and substance use disorders. Candidate biomarkers reviewed include various neuroimaging, genetic, molecular and peripheral assays, for the purposes of determining susceptibility or presence of illness, and predicting treatment response or safety. This review highlights a critical gap in the biomarker validation process. An enormous societal investment over the past 50 years has identified numerous candidate biomarkers. However, to date, the overwhelming majority of these measures have not been proven sufficiently reliable, valid and useful to be adopted clinically. It is time to consider whether strategic investments might break this impasse, focusing on a limited number of promising candidates to advance through a process of definitive testing for a specific indication. Some promising candidates for definitive testing include the N170 signal, an event-related brain potential measured using electroencephalography, for subgroup identification within autism spectrum disorder; striatal resting-state functional magnetic resonance imaging (fMRI) measures, such as the striatal connectivity index (SCI) and the functional striatal abnormalities (FSA) index, for prediction of treatment response in schizophrenia; error-related negativity (ERN), an electrophysiological index, for prediction of first onset of generalized anxiety disorder, and resting-state and structural brain connectomic measures for prediction of treatment response in social anxiety disorder. Alternate forms of classification may be useful for conceptualizing and testing potential biomarkers. Collaborative efforts allowing the inclusion of biosystems beyond genetics and neuroimaging are needed, and online remote acquisition of selected measures in a naturalistic setting using mobile health tools may significantly advance the field. Setting specific benchmarks for well-defined target application, along with development of appropriate funding and partnership mechanisms, would also be crucial. Finally, it should never be forgotten that, for a biomarker to be actionable, it will need to be clinically predictive at the individual level and viable in clinical settings.

Comparing the stability and reproducibility of brain-behaviour relationships found using Canonical Correlation Analysis and Partial Least Squares within the ABCD Sample

Introduction

Canonical Correlation Analysis (CCA) and Partial Least Squares Correlation (PLS) detect associations between two data matrices based on computing a linear combination between the two matrices (called latent variables; LVs). These LVs maximize correlation (CCA) and covariance (PLS). These different maximization criteria may render one approach more stable and reproducible than the other when working with brain and behavioural data at the population-level. This study compared the LVs which emerged from CCA and PLS analyses of brain-behaviour relationships from the Adolescent Brain Cognitive Development (ABCD) dataset and examined their stability and reproducibility.

Methods

Structural T1-weighted imaging and behavioural data were accessed from the baseline Adolescent Brain Cognitive Development dataset (N > 9000, ages = 9-11 years). The brain matrix consisted of cortical thickness estimates in different cortical regions. The behavioural matrix consisted of 11 subscale scores from the parent-reported Child Behavioral Checklist (CBCL) or 7 cognitive performance measures from the NIH Toolbox. CCA and PLS models were separately applied to the brain-CBCL analysis and brain-cognition analysis. A permutation test was used to assess whether identified LVs were statistically significant. A series of resampling statistical methods were used to assess stability and reproducibility of the LVs.

Results

When examining the relationship between cortical thickness and CBCL scores, the first LV was found to be significant across both CCA and PLS models (singular value: CCA = .13, PLS = .39, p < .001). LV1 from the CCA model found that covariation of CBCL scores was linked to covariation of cortical thickness. LV1 from the PLS model identified decreased cortical thickness linked to lower CBCL scores. There was limited evidence of stability or reproducibility of LV1 for both CCA and PLS. When examining the relationship between cortical thickness and cognitive performance, there were 6 significant LVs for both CCA and PLS (p < .01). The first LV showed similar relationships between CCA and PLS and was found to be stable and reproducible (singular value: CCA = .21, PLS = .43, p < .001).

Conclusion

CCA and PLS identify different brain-behaviour relationships with limited stability and reproducibility when examining the relationship between cortical thickness and parent-reported behavioural measures. However, both methods identified relatively similar brain-behaviour relationships that were stable and reproducible when examining the relationship between cortical thickness and cognitive performance. The results of the current study suggest that stability and reproducibility of brain-behaviour relationships identified by CCA and PLS are influenced by characteristics of the analyzed sample and the included behavioural measurements when applied to a large pediatric dataset.

Machine learning with neuroimaging biomarkers: Application in the diagnosis and prediction of drug addiction

Drug abuse is a serious problem worldwide. Owing to intermittent intake of certain substances and the early inconspicuous clinical symptoms, this brings huge challenges for timely diagnosing addiction status and preventing substance use disorders (SUDs). As a non-invasive technique, neuroimaging can capture neurobiological signatures of abnormality in multiple brain regions caused by drug consumption in each clinical stage, like parenchymal morphology alteration as well as aberrant functional activity and connectivity of cerebral areas, making it realizable to diagnosis, prediction and even preemptive therapy of addiction. Machine learning (ML) algorithms primarily used for classification have been extensively applied in analysing medical imaging datasets. Significant neurobiological characteristics employed and revealed by classifiers were used to diagnose addictive states and predict initiation and vulnerability to drug usage, treatment abstinence, relapse and resilience of addicts and the risk of SUD. In this review, we summarize application of ML methods in neuroimaging focusing on addicts' diagnosis of clinical status and risk prediction and elucidate the discriminative neurobiological features from brain electrophysiological, morphological and functional perspectives that contribute most to the classifier, finally highlighting the auxiliary role of ML in addiction treatment.

Threats to external validity in the neuroprediction of substance use treatment outcomes

Background: Tools predicting individual relapse risk would invaluably inform clinical decision-making (e.g. level-of-care) in substance use treatment. Studies of neuroprediction - use of neuromarkers to predict individual outcomes - have the dual potential to create such tools and inform etiological models leading to new treatments. However, financial limitations, statistical power demands, and related factors encourage restrictive selection criteria, yielding samples that do not fully represent the target population. This problem may be further compounded by a lack of statistical optimism correction in neuroprediction research, resulting in predictive models that are overfit to already-restricted samples.Objectives: This systematic review aims to identify potential threats to external validity related to restrictive selection criteria and underutilization of optimism correction in the existing neuroprediction literature targeting substance use treatment outcomes.Methods: Sixty-seven studies of neuroprediction in substance use treatment were identified and details of sample selection criteria and statistical optimism correction were extracted.Results: Most publications were found to report restrictive selection criteria (e.g. excluding psychiatric (94% of publications) and substance use comorbidities (69% of publications)) that would rule-out a considerable portion of the treatment population. Furthermore, only 21% of publications reported optimism correction.Conclusion: Restrictive selection criteria and underutilization of optimism correction are common in the existing literature and may limit the generalizability of identified neural predictors to the target population whose treatment they would ultimately inform. Greater attention to the inclusivity and generalizability of addiction neuroprediction research, as well as new opportunities provided through open science initiatives, have the potential to address this issue.

Task-based co-activation patterns reliably predict resting state canonical network engagement during development

Neurodevelopmental research has traditionally focused on development of individual structures, yet multiple lines of evidence indicate parallel development of large-scale systems, including canonical neural networks (i.e., default mode, frontoparietal). However, the relationship between region- vs. network-level development remains poorly understood. The current study tests the ability of a recently developed multi-task coactivation matrix approach to predict canonical resting state network engagement at baseline and at two-year follow-up in a large and cohort of young adolescents. Pre-processed tabulated neuroimaging data were obtained from the Adolescent Brain and Cognitive Development (ABCD) study, assessing youth at baseline (N = 6073, age = 10.0 ± 0.6 years, 3056 female) and at two-year follow-up (N = 3539, age = 11.9 ± 0.6 years, 1726 female). Individual multi-task co-activation matrices were constructed from the beta weights of task contrasts from the stop signal task, the monetary incentive delay task, and emotional N-back task. Activation-based predictive modeling, a cross-validated machine learning approach, was adopted to predict resting-state canonical network engagement from multi-task co-activation matrices at baseline. Note that the tabulated data used different parcellations of the task fMRI data ("ASEG" and Desikan) and the resting-state fMRI data (Gordon). Despite this, the model successfully predicted connectivity within the default mode network (DMN, rho = 0.179 ± 0.002, p < 0.001) across participants and identified a subset of co-activations within parietal and occipital macroscale brain regions as key contributors to model performance, suggesting an underlying common brain functional architecture across cognitive domains. Notably, predictive features for resting-state connectivity within the DMN identified at baseline also predicted DMN connectivity at two-year follow-up (rho = 0.258). These results indicate that multi-task co-activation matrices are functionally meaningful and can be used to predict resting-state connectivity. Interestingly, given that predictive features within the co-activation matrices identified at baseline can be extended to predictions at a future time point, our results suggest that task-based neural features and models are valid predictors of resting state network level connectivity across the course of development. Future work is encouraged to verify these findings with more consistent parcellations between task-based and resting-state fMRI, and with longer developmental trajectories.

Neural substrates of substance use disorders

PURPOSE OF REVIEW

Substance use disorders account for a tremendous burden to society, yet despite substantial progress in basic studies, our understanding of the brain-basis of these disorders is still emerging. This review summarizes the recent findings of neuroimaging studies with substance use disorder individuals.

RECENT FINDINGS

Resting-state functional connectivity studies support for some but not all substances of abuse and disruption in executive control. Structural neuroimaging findings point towards reduced subcortical volumes, which may emerge as an interaction between preexisting factors and recent substance use. Longitudinal studies implicate some of the same core brain structures and their functional role that have also been identified via case-control studies. Finally, meta-analyses support the idea of dysregulation of cortical control over subcortical salience processing.

SUMMARY

Although progress has been made and there is both structural and functional imaging evidence of an imbalance between brain structures involved in executive control and salience processing, there is emerging evidence that brain-behaviour relationships, which are core to discovering the neural processes that lead to and maintain substance use, are small and require larger consortia that prospectively examine individuals with substance use disorder.

Predicting the future of neuroimaging predictive models in mental health

Predictive modeling using neuroimaging data has the potential to improve our understanding of the neurobiology underlying psychiatric disorders and putatively information interventions. Accordingly, there is a plethora of literature reviewing published studies, the mathematics underlying machine learning, and the best practices for using these approaches. As our knowledge of mental health and machine learning continue to evolve, we instead aim to look forward and "predict" topics that we believe will be important in current and future studies. Some of the most discussed topics in machine learning, such as bias and fairness, the handling of dirty data, and interpretable models, may be less familiar to the broader community using neuroimaging-based predictive modeling in psychiatry. In a similar vein, transdiagnostic research and targeting brain-based features for psychiatric intervention are modern topics in psychiatry that predictive models are well-suited to tackle. In this work, we target an audience who is a researcher familiar with the fundamental procedures of machine learning and who wishes to increase their knowledge of ongoing topics in the field. We aim to accelerate the utility and applications of neuroimaging-based predictive models for psychiatric research by highlighting and considering these topics. Furthermore, though not a focus, these ideas generalize to neuroimaging-based predictive modeling in other clinical neurosciences and predictive modeling with different data types (e.g., digital health data).

Interpreting Brain Biomarkers: Challenges and solutions in interpreting machine learning-based predictive neuroimaging

Predictive modeling of neuroimaging data (predictive neuroimaging) for evaluating individual differences in various behavioral phenotypes and clinical outcomes is of growing interest. However, the field is experiencing challenges regarding the interpretability of the results. Approaches to defining the specific contribution of functional connections, regions, or networks in prediction models are urgently needed, which may help explore the underlying mechanisms. In this article, we systematically review the methods and applications for interpreting brain signatures derived from predictive neuroimaging based on a survey of 326 research articles. Strengths, limitations, and the suitable conditions for major interpretation strategies are also deliberated. In-depth discussion of common issues in existing literature and the corresponding recommendations to address these pitfalls are provided. We highly recommend exhaustive validation on the reliability and interpretability of the biomarkers across multiple datasets and contexts, which thereby could translate technical advances in neuroimaging into concrete improvements in precision medicine.

Identification and Validation of Distinct Latent Neurodevelopmental Profiles in the Adolescent Brain and Cognitive Development Study

BACKGROUND

Regardless of the precise mechanism, all neurodevelopmental models of risk assume that, at the population level, there exist subgroups of individuals that share similar patterns of neural function and development-and that these subgroups somehow relate to psychiatric risk. However, the existence of multiple neurodevelopmental subgroups at the population level has not been assessed previously.

METHODS

In the current study, cross-validated latent profile analysis was used to test for the presence of empirically derived, brain-based developmental subgroups using functional magnetic resonance imaging data from 6758 individuals (49.4% female; mean age = 9.94 years) in the Adolescent Brain and Cognitive Development (ABCD) study wave 1 release. Data were randomly split into training and testing samples.

RESULTS

Analyses in the training sample (n = 3379) identified a seven-profile solution (entropy = 0.880) that was replicated in the held-out testing data (n = 3379, entropy = 0.890). Identified subgroups included a moderate group (66.8%), high reward (4.3%) and low reward (4.0%) groups, high inhibition (9.8%) and low inhibition (6.7%) groups, and high emotion regulation (4.0%) and low emotion regulation (4.3%) groups. Relative to the moderate group, other subgroups were characterized by more males (χ2 = 24.10, p = .0005), higher proportions of individuals from lower-income households (χ2 = 122.17, p < .0001), poorer cognitive performance (ps < .0001), more screen time (F = 6.80, p < .0001), heightened impulsivity (ps < .006), and higher rates of neurodevelopmental disorders (χ2 = 26.20, p = .0002).

CONCLUSIONS

These data demonstrate the existence of multiple, distinct neurodevelopmental subgroups at the population level. They indicate that these empirically derived, brain-based developmental profiles relate to differences in clinical features, even at a young age, and prior to the peak period of risk for the development of psychopathology.

Large-scale functional brain networks of maladaptive childhood aggression identified by connectome-based predictive modeling

Disruptions in frontoparietal networks supporting emotion regulation have been long implicated in maladaptive childhood aggression. However, the association of connectivity between large-scale functional networks with aggressive behavior has not been tested. The present study examined whether the functional organization of the connectome predicts severity of aggression in children. This cross-sectional study included a transdiagnostic sample of 100 children with aggressive behavior (27 females) and 29 healthy controls without aggression or psychiatric disorders (13 females). Severity of aggression was indexed by the total score on the parent-rated Reactive-Proactive Aggression Questionnaire. During fMRI, participants completed a face emotion perception task of fearful and calm faces. Connectome-based predictive modeling with internal cross-validation was conducted to identify brain networks that predicted aggression severity. The replication and generalizability of the aggression predictive model was then tested in an independent sample of children from the Adolescent Brain Cognitive Development (ABCD) study. Connectivity predictive of aggression was identified within and between networks implicated in cognitive control (medial-frontal, frontoparietal), social functioning (default mode, salience), and emotion processing (subcortical, sensorimotor) (r = 0.31, RMSE = 9.05, p = 0.005). Out-of-sample replication (p < 0.002) and generalization (p = 0.007) of findings predicting aggression from the functional connectome was demonstrated in an independent sample of children from the ABCD study (n = 1791; n = 1701). Individual differences in large-scale functional networks contribute to variability in maladaptive aggression in children with psychiatric disorders. Linking these individual differences in the connectome to variation in behavioral phenotypes will advance identification of neural biomarkers of maladaptive childhood aggression to inform targeted treatments.

Functional connectivity during frustration: a preliminary study of predictive modeling of irritability in youth

Irritability cuts across many pediatric disorders and is a common presenting complaint in child psychiatry; however, its neural mechanisms remain unclear. One core pathophysiological deficit of irritability is aberrant responses to frustrative nonreward. Here, we conducted a preliminary fMRI study to examine the ability of functional connectivity during frustrative nonreward to predict irritability in a transdiagnostic sample. This study included 69 youths (mean age = 14.55 years) with varying levels of irritability across diagnostic groups: disruptive mood dysregulation disorder (n = 20), attention-deficit/hyperactivity disorder (n = 14), anxiety disorder (n = 12), and controls (n = 23). During fMRI, participants completed a frustrating cognitive flexibility task. Frustration was evoked by manipulating task difficulty such that, on trials requiring cognitive flexibility, "frustration" blocks had a 50% error rate and some rigged feedback, while "nonfrustration" blocks had a 10% error rate. Frustration and nonfrustration blocks were randomly interspersed. Child and parent reports of the affective reactivity index were used as dimensional measures of irritability. Connectome-based predictive modeling, a machine learning approach, with tenfold cross-validation was conducted to identify networks predicting irritability. Connectivity during frustration (but not nonfrustration) blocks predicted child-reported irritability (ρ = 0.24, root mean square error = 2.02, p = 0.03, permutation testing, 1000 iterations, one-tailed). Results were adjusted for age, sex, medications, motion, ADHD, and anxiety symptoms. The predictive networks of irritability were primarily within motor-sensory networks; among motor-sensory, subcortical, and salience networks; and between these networks and frontoparietal and medial frontal networks. This study provides preliminary evidence that individual differences in irritability may be associated with functional connectivity during frustration, a phenotype-relevant state.

Interactions between methodological and interindividual variability: How Monetary Incentive Delay (MID) task contrast maps vary and impact associations with behavior

INTRODUCTION

Phenomena related to reward responsiveness have been extensively studied in their associations with substance use and socioemotional functioning. One important task in this literature is the Monetary Incentive Delay (MID) task. By cueing and delivering performance-contingent reward, the MID task has been demonstrated to elicit robust activation of neural circuits involved in different phases of reward responsiveness. However, systematic evaluations of common MID task contrasts have been limited to between-study comparisons of group-level activation maps, limiting their ability to directly evaluate how researchers' choice of contrasts impacts conclusions about individual differences in reward responsiveness or brain-behavior associations.

METHODS

In a sample of 104 participants (Age Mean = 19.3, SD = 1.3), we evaluate similarities and differences between contrasts in: group- and individual-level activation maps using Jaccard's similarity index, region of interest (ROI) mean signal intensities using Pearson's r, and associations between ROI mean signal intensity and psychological measures using Bayesian correlation.

RESULTS

Our findings demonstrate more similarities than differences between win and loss cues during the anticipation contrast, dissimilarity between some win anticipation contrasts, an apparent deactivation effect in the outcome phase, likely stemming from the blood oxygen level-dependent undershoot, and behavioral associations that are less robust than previously reported.

CONCLUSION

Consistent with recent empirical findings, this work has practical implications for helping researchers interpret prior MID studies and make more informed a priori decisions about how their contrast choices may modify results.

Brain Networks Associated With COVID-19 Risk: Data From 3662 Participants

Background

Our behavioral traits, and subsequent actions, could affect the risk of exposure to the coronavirus disease of 2019 (COVID-19). The current study aimed to determine whether unique brain networks are associated with the COVID-19 infection risk.

Methods

This research was conducted using the UK Biobank Resource. Functional magnetic resonance imaging scans in a cohort of general population (n = 3662) were used to compute the whole-brain functional connectomes. A network-informed machine learning approach was used to identify connectome and nodal fingerprints that are associated with positive COVID-19 status during the pandemic up to February fourth, 2021.

Results

The predictive models successfully identified 6 fingerprints that were associated with COVID-19 positive, compared to negative status (all p values < 0.005). Overall, lower integration across the brain modules and increased segregation, as reflected by internal within module connectivity, were associated with higher infection rates. More specifically, COVID-19 positive status was associated with 1) reduced connectivity between the central executive and ventral salience, as well as between the dorsal salience and default mode networks; 2) increased internal connectivity within the default mode, ventral salience, subcortical and sensorimotor networks; and 3) increased connectivity between the ventral salience, subcortical and sensorimotor networks.

Conclusion

Individuals are at increased risk of COVID-19 infections if their brain connectome is consistent with reduced connectivity in the top-down attention and executive networks, along with increased internal connectivity in the introspective and instinctive networks. These identified risk networks could be investigated as target for treatment of illnesses with impulse control deficits.

Transdiagnostic, Connectome-Based Prediction of Memory Constructs Across Psychiatric Disorders

Memory deficits are observed in a range of psychiatric disorders, but it is unclear whether memory deficits arise from a shared brain correlate across disorders or from various dysfunctions unique to each disorder. Connectome-based predictive modeling is a computational method that captures individual differences in functional connectomes associated with behavioral phenotypes such as memory. We used publicly available task-based functional MRI data from patients with schizophrenia (n = 33), bipolar disorder (n = 34), attention deficit hyper-activity disorder (n = 32), and healthy controls (n = 73) to model the macroscale brain networks associated with working, short- and long-term memory. First, we use 10-fold and leave-group-out analyses to demonstrate that the same macroscale brain networks subserve memory across diagnostic groups and that individual differences in memory performance are related to individual differences within networks distributed throughout the brain, including the subcortex, default mode network, limbic network, and cerebellum. Next, we show that diagnostic groups are associated with significant differences in whole-brain functional connectivity that are distinct from the predictive models of memory. Finally, we show that models trained on the transdiagnostic sample generalize to novel, healthy participants (n = 515) from the Human Connectome Project. These results suggest that despite significant differences in whole-brain patterns of functional connectivity between diagnostic groups, the core macroscale brain networks that subserve memory are shared.

Predicting alcohol dependence from multi-site brain structural measures

To identify neuroimaging biomarkers of alcohol dependence (AD) from structural magnetic resonance imaging, it may be useful to develop classification models that are explicitly generalizable to unseen sites and populations. This problem was explored in a mega‐analysis of previously published datasets from 2,034 AD and comparison participants spanning 27 sites curated by the ENIGMA Addiction Working Group. Data were grouped into a training set used for internal validation including 1,652 participants (692 AD, 24 sites), and a test set used for external validation with 382 participants (146 AD, 3 sites). An exploratory data analysis was first conducted, followed by an evolutionary search based feature selection to site generalizable and high performing subsets of brain measurements. Exploratory data analysis revealed that inclusion of case‐ and control‐only sites led to the inadvertent learning of site‐effects. Cross validation methods that do not properly account for site can drastically overestimate results. Evolutionary‐based feature selection leveraging leave‐one‐site‐out cross‐validation, to combat unintentional learning, identified cortical thickness in the left superior frontal gyrus and right lateral orbitofrontal cortex, cortical surface area in the right transverse temporal gyrus, and left putamen volume as final features. Ridge regression restricted to these features yielded a test‐set area under the receiver operating characteristic curve of 0.768. These findings evaluate strategies for handling multi‐site data with varied underlying class distributions and identify potential biomarkers for individuals with current AD.

Identification of a brain fingerprint for overweight and obesity

The brain plays a central role in the pathophysiology of overweight and obesity. Connectome-based Predictive Modeling (CPM) is a newly developed, data-driven approach that exploits whole-brain functional connectivity to predict a behavior or trait that varies across individuals. We used CPM to determine whether brain "fingerprints" evoked during milkshake consumption could be isolated for common measures of adiposity in 67 adults with overweight and obesity. We found that CPM captures more variance in waist circumference than either percent body fat or BMI, the most frequently used measures to assess brain correlates of obesity. In a post-hoc analysis, we were also able to derive a largely separable functional connectivity network predicting fasting blood insulin. These findings suggest that, in individuals with overweight and obesity, brain network patterns may be more tightly coupled to waist circumference than BMI or percent body fat and that adiposity and glucose tolerance are associated with distinct maps, pointing to dissociable central pathophysiological phenotypes for obesity and diabetes.